Hi all,

I've been trying to find an answer to my question by reading the posts on the Forum but I have not been able to find an answer, so that is why I am now writing my first ever post. I realize that my knowledge of Stata and statistics in general is very limited in comparison with the people here, that is why I need some help.

I am doing panel data analysis. I am looking at 21 stock listed football clubs and investigating what the effect of a win/draw/loss in a football match is on the abnormal return of their stock one day after the match. I have a lot of days (1557) and the panel data is highly unbalanced (because football clubs do not all play football on the same day).

In further models I also look at the effects of whether a win/draw/loss was surprising or expected and the effects of domestic league vs. European league matches, but for simplicity I will focus now on the most simple model:

AR1: the abnormal return on the first trading day after a match
W: dummy that =1 if the result was a win
L: dummy that =1 if the result was a loss
Thus, the constant term captures the effect of a draw

1) xtreg AR1 W L, fe

As I understand it, this regression takes care of the football club Fixed Effects, but not the time Fixed effects. To control for time Fixed Effects I add i.tradingday (my time variable):

2) xtreg AR1 W L i.tradingday, fe

I used testparm i.tradingdays to see if it is necessary to add the time dummies. Since the p-value = 0.0001 I conclude that adding the time dummies is necessary.

Comparing the results of 1) and 2) I find that the coefficients for W and L are very similar and highly significant in both cases. Virtually all the time dummies in 2) are insignificant.
However, the _cons coefficient (representing a draw) is highly significant in 1), but when adding the high amount of time dummies in 2) the _cons coefficient becomes insignificant.

Having a significant _cons coefficient is important to me for two reasons: 1. It tells me the effect of a draw (which in 1) is -0.8%***) and 2. I need to add up the _cons coefficient with the coefficients of W and L to calculate their respective effects.

I tried to change my model by using loss as a base case instead of draw:

3) xtreg AR1 W D, fe
4) xtreg AR1 W D i.tradingday, fe

Again, the coefficients for W and D are highly significant and similar in both cases, however the _cons coefficient (now representing a loss) becomes insignificant.

Why is it that the _cons coefficient becomes insignificant after controlling for time Fixed Effects?
What can I do about this?

Thanks in advance for any responses.